jnthn Heh...seems the JIT missing the new write barriers seems to be to blame for the SEGVs. 10:21
Won't be the only reason, but nqp's make test passes with JIT disabled :) 10:22
dalek arVM/reframe: 352293c | jnthn++ | src/jit/emit_x64.dasc:
Add new write barrier to bindlex JIT.
 10:36
jnthn Odd, some tests now explode in the debugger and under make test, but run to completion if run directly... 10:37
Oh, interesting. The JIT caches FRAME in a register but that can move if they're GC-able 10:39
dalek arVM/reframe: 1268c1b | jnthn++ | src/jit/emit_x64.dasc:
Fix a typo.
 10:40
masak meaning, the cache can hold the old address too long? 10:56
jnthn Aye
And it looks a tad tricky to fix
jnthn wonders if brrt might show up at some point today :)
masak wonders if there's a Java framework that purports to solve cache invalidation :) 10:57
jnthn Probably not for the case where the cache is *a CPU register*!
masak :P 11:00
jnthn This is just one of those refactors that ends up breaking something that's been widely used as an invariant.
So it needs changes in quite a lot of places. 11:01
nwc10 ==11156==ERROR: AddressSanitizer: SEGV on unknown address 0x000000000000 (pc 0x7f7dde8b91ef sp 0x7fffa6ac0340 bp 0x7fffa6ac2000 T0) #0 0x7f7dde8b91ee in MVM_interp_run src/core/interp.c:441 11:18
(still)
AFK - POETS
jnthn :) 11:34
masak is nwc10 going to meet some poets, or is he going to recite some poetry with poets, or is he being chased by poets? 11:53
jnthn It's an acronym ;) 11:56
masak .oO( Poets On Every Third Syllable ) 11:57
timotimo i thought it'd be about very loud poetry 12:02
is the frame forced to be in one of the callee-saved registers? 12:03
because if it's a caller-saved register, we'd already know exactly where we have to restore it, and we could just restore it from - i guess - tc->current_frame or whats-it-called each time 12:04
diakopter masak: I thought he was hungry for poets 14:44
masak maybe he was a stone lion eating poet 15:28
TimToady notes that 'lion' and 'poet' are both four-letter words, but then again, so are 'four' and 'word' 16:04
timotimo letter ought to be a one-letter-word :( 16:05
then again, 'letter' eq 'letter' x 1
so i suppose letter *is* a one-'letter'-word
TimToady sophists, the lot of you :) 16:06
timotimo i think i actually might be an asophist 16:08

Powered by Cro and the Raku® Programming Language.

Please report any issues / comments / feature requests as an issue on App::Raku::Log.

Thank you!